Article
Biochemical Research Methods
Zhichao Wang, Xuelei Li, Jianping Fan, Jintao Meng, Zhenli Lin, Yi Pan, Yanjie Wei
Summary: Spiking neural network (SNN) simulators are vital in modeling neural systems and studying brain functions. Our developed SNN simulator SWsnn, based on the new Sunway SW26010pro processor, runs faster and shows strong performance advantages in simulating neural networks.
JOURNAL OF COMPUTATIONAL BIOLOGY
(2023)
Article
Computer Science, Software Engineering
Jingwei Huang, Shanshan Zhang, Bo Duan, Yanfeng Zhang, Xiaoyang Guo, Mingwei Sun, Li Yi
Summary: We propose a novel vectorized indoor modeling approach that converts point clouds into concise and semantically segmented polygonal meshes of building information models (BIM). Our method ArrangementNet leverages co-linear and co-face relationships in the arrangement to estimate scene arrangements from incomplete point clouds and improves the quality of BIM model reconstruction.
ACM TRANSACTIONS ON GRAPHICS
(2023)
Article
Engineering, Biomedical
Yifei Feng, Shijia Geng, Jianjun Chu, Zhaoji Fu, Shenda Hong
Summary: This paper introduces a deep spiking neural network (SNN) for ECG classification, and compares the effects of different ANN activation functions on SNN performance.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2022)
Article
Computer Science, Artificial Intelligence
Yikang Yang, Jia Ren, Feng Duan
Summary: The researchers proposed a method using spiking neural network for recognizing sEMG signals, addressing the issues of information loss and poor training performance. They converted sEMG to spike trains using a smoothed frequency-domain decomposition encoder and converted network output to recognition results using a network efferent energy decoder. The method was validated in a hand gestures recognition task, demonstrating high accuracy.
COGNITIVE COMPUTATION
(2023)
Article
Computer Science, Artificial Intelligence
Fatemehossadat Miri, Carol Miles, Harold W. Lewis
Summary: Tritonia escape swim network and its neural activities have been studied in the laboratory, leading to significant advances in identifying the biological components involved. Neuronal patterns of the escape swim circuit have been artificially reproduced, including interneurons and other participating neurons. Spike patterns of each neuron were successfully simulated and validated using a spiking neural network simulator.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Neurosciences
Yulong Yan, Haoming Chu, Yi Jin, Yuxiang Huan, Zhuo Zou, Lirong Zheng
Summary: The article proposes a sparsity-driven spiking neural network learning algorithm, which aims to achieve improved spiking and synaptic sparsity by utilizing backpropagation with spiking regularization. Experimental results show that the algorithm achieves good accuracy and synaptic sparsity on multiple datasets.
FRONTIERS IN NEUROSCIENCE
(2022)
Article
Computer Science, Artificial Intelligence
Tingfang Wu, Qiang Lyu, Linqiang Pan
Summary: This study explores spiking neural P systems (SNP systems) and their variant evolution-communication SNP (ECSNP) systems, demonstrating the Turing universality of ECSNP systems as number-generating devices and highlighting the critical impact of the quantity of spikes in neurons on the computational power of ECSNP systems.
INTERNATIONAL JOURNAL OF NEURAL SYSTEMS
(2021)
Article
Automation & Control Systems
Chenglong Zou, Xiaoxin Cui, Guang Chen, Yuanyuan Jiang, Yuan Wang
Summary: A novel artificial neural network to spiking neural network (ANN-to-SNN) conversion framework is presented for high-accuracy and low-latency SNNs with negative-spike dynamics. The conversion algorithm takes median quantization constraint and spike compensation technique into consideration and achieves truly lossless accuracy performance. In addition, the spiking models need quite shortened computing time steps among other works and consume much fewer synaptic operations than their ANN counterparts.
ADVANCED INTELLIGENT SYSTEMS
(2023)
Article
Computer Science, Information Systems
Ziyu Jia, Junyu Ji, Xinliang Zhou, Yuhan Zhou
Summary: Automatic sleep staging is crucial for assessing sleep quality, and this study introduces a new hybrid spiking neural network model for this purpose. By utilizing a spiking neural network and a hybrid back propagation algorithm, the proposed method demonstrates satisfactory performance on the ISRUC-SLEEP dataset.
SCIENCE CHINA-INFORMATION SCIENCES
(2022)
Article
Engineering, Electrical & Electronic
Qinyu Chen, Guoqiang He, Xinyuan Wang, Jin Xu, Sirui Shen, Hui Chen, Yuxiang Fu, Li Li
Summary: Spiking Neural Networks (SNNs) are a promising technology for low power and high-performance edge computing hardware design. Compared to traditional Artificial Neural Networks (ANNs), SNNs provide more realistic brain-inspired computing models and serve as an alternative to ANNs. However, the temporal characteristics of SNNs lead to irregular and repeated data accesses, resulting in increased latency and power consumption. This paper proposes an efficient architecture for SNNs that utilizes event-based characteristics, a reconfigurable spiking neuron processing unit, and an efficient dataflow with fast-filtering mechanism to reduce the required cycles per frame.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2022)
Article
Neurosciences
Dengyu Wu, Xinping Yi, Xiaowei Huang
Summary: This article discusses the trend of developing energy-efficient Spiking Neural Networks using a well-trained Convolutional Neural Network (CNN) as the basis. The proposed framework ensures equalization between the output of CNN and the accumulated current in SNN, improving accuracy, latency, and energy efficiency. Experimental results on various neural network architectures demonstrate the superior performance of the proposed method.
FRONTIERS IN NEUROSCIENCE
(2022)
Article
Computer Science, Information Systems
Kwanghyun Koo, Hyun Kim
Summary: In this study, a new vectorized structured kernel pruning method is proposed, which achieves high FLOPs reduction and minimal accuracy degradation while maintaining the weight structure. Experimental results demonstrate significant parameter and FLOPs reduction, as well as real acceleration effects on GPUs, in various networks including ResNet-50.
Article
Computer Science, Artificial Intelligence
Yongcheng Zhou, Anguo Zhang
Summary: The study shows that using the SynSup and SynAcc mechanisms can effectively improve the inference speed of spiking neural networks compared to the general I&F model.
APPLIED INTELLIGENCE
(2021)
Article
Chemistry, Multidisciplinary
Amirhossein Javanshir, Thanh Thi Nguyen, M. A. Parvez Mahmud, Abbas Z. Kouzani
Summary: This paper proposes a novel metaheuristic-based supervised learning method for spiking neural networks (SNNs) to overcome the challenges of training SNNs with backpropagation-based supervised learning methods. Experimental results show that the proposed algorithm outperforms other experimental algorithms in solving four classification benchmark datasets, with Cuckoo Search (CS) reporting the best performance.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Information Systems
Donghyung Yoo, Doo Seok Jeong
Summary: This study proposes a novel alternative to convolution in spiking neural networks (SNNs) called SNNs with trainable dynamic time-surfaces (DTS-SNNs), aiming to improve computational efficiency. Through evaluation on real-world event-based datasets, the results demonstrate high classification accuracies and significant improvements in computational efficiency for DTS-SNNs.
Article
Computer Science, Software Engineering
Jonathan X. Zheng, Samraat Pawar, Dan F. M. Goodman
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2019)
Article
Neurosciences
Charlotte Le Mouel, Romain Tisserand, Thomas Robert, Romain Brette
Article
Multidisciplinary Sciences
Marcel Stimberg, Dan F. M. Goodman, Thomas Nowotny
SCIENTIFIC REPORTS
(2020)
Article
Biology
Anirudh Kulkarni, Irene Elices, Nicolas Escoubet, Lea-Laetitia Pontani, Alexis Michel Prevost, Romain Brette
JOURNAL OF EXPERIMENTAL BIOLOGY
(2020)
Article
Biology
Titipat Achakulvisut, Tulakan Ruangrong, Isil Bilgin, Sofie Van Den Bossche, Brad Wyble, Dan F. M. Goodman, Konrad P. Kording
Article
Biology
Sarah Goethals, Romain Brette
Article
Multidisciplinary Sciences
Mickael Zbili, Sylvain Rama, Pierre Yger, Yanis Inglebert, Norah Boumedine-Guignon, Laure Fronzaroli-Moliniere, Romain Brette, Michael Russier, Dominique Debanne
Article
Computer Science, Software Engineering
Jonathan X. Zheng, Samraat Pawar, Dan F. M. Goodman
Summary: The study presents an algorithm for constructing confluent drawings by solving the node split and short-circuit problems to maintain the hierarchical structure of power groups, classifying the resulting drawings as 'power-confluent'. The research also includes an improved method for power graph construction.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2021)
Article
Neurosciences
Yaqing Su, Yoojin Chung, Dan F. M. Goodman, Kenneth E. Hancock, Bertrand Delgutte
Summary: The study revealed neural mechanisms differences in pitch perception of irregular sounds between rabbits with normal hearing and those with cochlear implants. IC neurons showed tuning of firing rate to average pulse rate and differences in synchronized responses to periodic and irregular pulse trains in both groups of rabbits.
JARO-JOURNAL OF THE ASSOCIATION FOR RESEARCH IN OTOLARYNGOLOGY
(2021)
Review
Neurosciences
Romain Brette
Summary: Paramecium is a unicellular organism that swims in fresh water by beating cilia. When stimulated, it exhibits avoiding reaction triggered by a calcium-based action potential. Some authors have referred to Paramecium as a swimming neuron.
Article
Neurosciences
Sarah Goethals, Martijn C. Sierksma, Xavier Nicol, Annabelle Reaux-Le Goazigo, Romain Brette
Summary: The study measured the axial current produced by the axon initial segment of mouse retinal ganglion cells, finding it to be large, requiring high sodium channel conductance density, and covarying with cell capacitance to depolarize the cell by approximately 30mV. During sustained depolarization, the current attenuated but broadened temporally to preserve somatic depolarization. This suggests that the properties of the initial segment are adjusted to ensure reliable backpropagation of the axonal action potential.
JOURNAL OF NEUROPHYSIOLOGY
(2021)
Article
Multidisciplinary Sciences
Aurelie Fekete, Norbert Ankri, Romain Brette, Dominique Debanne
Summary: The distal shift of the axon initial segment (AIS) position increases axial resistance and excitability in L-5 pyramidal neurons, resulting in a decrease in the voltage threshold of the somatic action potential.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2021)
Article
Multidisciplinary Sciences
Nicolas Perez-Nieves, Vincent C. H. Leung, Pier Luigi Dragotti, Dan F. M. Goodman
Summary: The authors demonstrate that heterogeneity in spiking neural networks can improve task performance, stability, and robustness, especially for tasks with rich temporal structures. Additionally, the distribution of neuronal parameters in the trained networks is similar to those observed experimentally, suggesting that heterogeneity may play an active role in helping animals adapt to changing environments in the brain.
NATURE COMMUNICATIONS
(2021)
Article
Multidisciplinary Sciences
Nicolas Escoubet, Romain Brette, Lea-Laetitia Pontani, Alexis Michel Prevost
Summary: In this study, we investigated the swimming behavior of Paramecium tetraurelia, a unicellular microorganism, in micro-engineered pools with obstacles. We measured two types of contact interactions, passive scattering and avoiding reactions (ARs). We found that ARs were only mechanically triggered 10% of the time and that two-thirds of the ARs had a delay of approximately 150 ms.
ROYAL SOCIETY OPEN SCIENCE
(2023)
Article
Acoustics
Isaac Engel, Dan F. M. Goodman, Lorenzo Picinali
Summary: Binaural rendering of Ambisonics signals is a common method for reproducing spatial audio content. In this study, nine HRTF preprocessing methods were compared, and it was found that certain methods performed better at lower spatial orders. A newly proposed method, BiMagLS, displayed the best overall performance and is recommended for bilateral Ambisonics signal rendering. The results indirectly validate the ability of perceptual models to predict listeners' responses.
